I am not an ML person by all means, but when I played with GPUs, I found that the time it takes to get something done using CUDA is substantially lower than wit

I am not an ML person by all means, but when I played with GPUs, I found that the time it takes to get something done using CUDA is substantially lower than with OpenCL, nota bene I have not yet tried Metal. The barrier to entry also seems to be lower for CUDA, so this might be something that TensorFlow people are considering important.